Introduction
Recent advances in photonic integration technologies[1] and theoretical progress in optical computing architectures[2], [3], have fueled research interest in photonic neuromorphic computing, aiming to transfer the low-power and high-speed and density credentials of light manipulating components in parallel computing layouts[4]–[6]. Previous implementations of photonic neuromorphic hardware can be classified in two broad categories: i) in multiwavelength resonant layouts, where every axon is realized via a different wavelength[7]–[11], ii) in coherent layouts that use a single laser source and utilize the phase of transmitted light both for sign representation as well as for matrix manipulation in interferometric layouts[12]–[14].